Asset Prices in an Exchange Economy

Author(s): Robert E. Lucas, Jr.

Source: Econometrica, Vol. 46, No. 6 (Nov., 1978), pp. 1429-1445
Published by: The Econometric Society
Stable URL: http://www.jstor.org/stable/1913837
Accessed: 08/06/2010 21:38
Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at
http://www.jstor.org/page/info/about/policies/terms.jsp. JSTOR's Terms and Conditions of Use provides, in part, that unless
you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you
may use content in the JSTOR archive only for your personal, non-commercial use.
Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at
http://www.jstor.org/action/showPublisher?publisherCode=econosoc.
Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed
page of such transmission.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.
The Econometric Society is collaborating with JSTOR to digitize, preserve and extend access to Econometrica.
http://www.jstor.org
Econometrica, Vol. 46, No. 6 (November, 1978)
ASSET PRICES IN AN EXCHANGE ECONOMY
BY ROBERT E. LUCAS, JR.'
This paper is a theoretical examination of the stochastic behavior of equilibrium asset
prices in a one-good, pure exchange economy with identical consumers. A general
method of constructing equilibrium prices is developed and applied to a series of
examples.
1. INTRODUCTION
THIS PAPER IS A THEORETICAL examination of the stochastic behavior of equi-
librium asset prices in a one-good, pure exchange economy with identical
consumers. The single good in this economy is (costlessly) produced in a number
of different productive units; an asset is a claim to all or part of the output of one
of these units. Productivity in each unit fluctuates stochastically through time, so
that equilibrium asset prices will fluctuate as well. Our objective will be to
understand the relationship between these exogenously determined productivity
changes and market determined movements in asset prices.
Most of our attention will be focused on the derivation and application of a
functional equation in the vector of equilibrium asset prices, which is solved for
price as a function of the physical state of the economy. This equation is a
generalization of the Martingale property of stochastic price sequences, which
serves in practice as the defining characteristic of market "efficiency," as that
term is used by Fama [7] and others. The model thus serves as a simple context
for examining the conditions under which a price series' failure to possess the
Martingale property can be viewed as evidence of non-competitive or "irra-
tional" behavior.
The analysis is conducted under the assumption that, in Fama's terms, prices
"fully reflect all available information," an hypothesis which Muth [13] had
earlier termed "rationality of expectations." As Muth made clear, this hypoth-
esis (like utility maximization) is not "behavioral": it does not describe the way
agents think about their environment, how they learn, process information, and
so forth. It is rather a property likely to be (approximately) possessed by the
outcome of this unspecified process of learning and adapting. One would feel
more comfortable, then, with rational expectations equilibria if these equilibria
were accompanied by some form of "stability theory" which illuminated the
forces which move an economy toward equilibrium. The present paper also
offers a convenient context for discussing this issue.
The conclusions of this paper with respect to the Martingale property precisely
replicate those reached earlier by LeRoy (in [10] and [11]), and not surprisingly,
since the economic reasoning in [10] and the present paper is the same. The
lThis paper originated in a conversation with Pentti Kouri, who posed to me the problem studied
below. I would also like to thank Yehuda Freidenberg, Jose Scheinkman, and Joseph Williams for
many helpful comments.
1429
1430 ROBERT E. LUCAS, JR.
context used here differs somewhat from LeRoy's, however, and the analytical
methods used differ considerably.
The economy is informally described in the next section, and equilibrium is
formally defined in Section 3. In Section 4, the basic functional equation for
prices is derived and studied. Section 5 develops a certain "duality" property, on
which is based the discussion of stability in Section 6. Section 7 deals with
examples which are simple enough to permit either explicit solution or some
"comparative static" exercises. The role of the Martingale property is discussed
in Section 8. Section 9 concludes the paper.
2. DESCRIPTION OF THE ECONOMY
Consider an economy with a single consumer, interpreted as a representative
"stand in" for a large number of identical consumers. He wishes to maximize the
quantity
(1) E1 E 3tU(ct)}
where
c,
is a stochastic process representing consumption of a single good, ,B is a
discount factor, U() is a current period utility function, and E{ } is an expec-
tations operator.
The consumption good is produced on n distinct productive units. Let Yit be
the output of unit i in period t, i =
1, ..., n, and let
yt
=
(yIt,
.... Ynt) be the
output vector in t. Output is perishable, so that feasible consumption levels are
those which satisfy
Production is entirely "exogenous": no resources are utilized, and there is no
possibility of affecting the output of any unit at any time. The motion of
yt
will be
taken to follow a Markov process, defined by its transition function
F(y', y)= pr
IYt+?
:
Y'lYt
= Y}.
Ownership in these productive units is determined each period in a competi-
tive stock market. Each unit has outstanding one perfectly divisible equity share.
A share entitles its owner as of the beginning of t to all of the unit's output in
period t. Shares are traded, after payment of real dividends, at a competitively
determined price vector Pt
=
(PIt
. . .
,
Pnt). Let
zt
=
(Zlt.
. . Znt)
denote a
consurner's beginning-of-period share holdings.
In this economy, it is easy to determine equilibrium quantities of consumption
and asset holdings. All output will be consumed
(ct
=
:iyit)
and all shares will be
held
(z, (1,
1,
. . . 1)=
1 for all
t).
The main
analytical issue, then, will be the
determination of equilibrium price behavior.
Our attack on this problem begins from the observation that all relevant
information on the current and future physical state of the economy is sum-
ASSET PRICES 1431
marized in the current output vector y. Since, given recursive preferences, the
asset market "solves" a problem of the same form each period, equilibrium
prices should (if they behave in a systematic way at all) be expressible as some
fixed function p( ) of the state of the economy, or
p,= p(y,)
where the ith
coordinate
pi(y,)
is the price of a share of unit i when the economy is in the state
y,.
If
so, knowledge
of the transition function
F(y', y)
and this function
p(y)
will
suffice to determine the stochastic character of the price process
{p,}.
Similarly, one would expect a consumer's current consumption and portfolio
decisions, c, and
z,,1,
to depend on his beginning of period portfolio, z, the
prices he faces,
p,,
and the relevant information he possesses on current and
future states of the economy,
y,
If so, his behavior can be described by fixed
decision rules c(-) and g(-): c,
=
c(z,, yt, pt)
and
zt+1
=
g(zt, Yt, Pt).
Now given perceived, future price behavior F(y', y) and p(y), consumers will
be able to determine these decision rules c( ) and g( ) optimally. In this sense, a
price function p determines consumer behavior. On the other hand, given
decision rules c( ) and g( ), the current period market clearing conditions
determine a price function p( ). In this sense, consumer behavior determines the
equilibrium price function. We close the system with the assumption of rational
expectations: the market clearing price function p implied by consumer behavior
is assumed to be the same as the price function p on which consumer decisions
are based.
3. DEFINITION OF EQUILIBRIUM
The economy described in the preceding section is specified by the functions U
and F and the number ,B. Assume 0 < 3 < 1. U: R
+
--
R
+
is continuously
differentiable, bounded, increasing, and strictly concave, with U(0)= 0.'
F: E`'+ x
En
R is continuous; F(-, y)
is a distribution function for each fixed y,
with F(O, y)= 0. Assume that the process defined by F has a stationary dis-
tribution f(-), the unique solution to
(y')
F
E(y',y)dq (y),
and that for any continuous function g(y),
J g(y') dF
(y',y)
is a continuous function of y.
An equilibrium will be a pair of functions: a price function p(y), as discussed
above, and an optimum value function v(z, y). The value v(z, y)
will be inter-
preted as the value of the objective (1) for a consumer who begins in state y with
holdings z, and follows an optimum consumption-portfolio policy thereafter.
2Rt is the set of nonnegative real numbers. En is n-dimensional space. E" is the subset of
En
with all components nonnegative (xE EE and x
-
0). LU is the set of continuous, bounded functions
with domain En, and so on.
1432 ROBERT E. LUCAS, JR.
DEFINITION: An equilibrium is a continuous function p(y): E+- E'+ and a
continuous, bounded function v (z, y): E'+ x E' +
-
R + such that
(i)
ttv(z,Y)=max U(c)+,8{v(x,y')dF((y',y)
c,x
subject to
c +p(y) x S y z +p(y) z, c BO, O!x !z,
where z is a vector with components exceeding one;
(ii) for each y, v(i, y) is attained by c
=
Eiyi
and x = 1.
Condition (i) says that, given the behavior of prices, a consumer allocates his
resources y z+p(y) z optimally among current consumption c and end-of-
period share holdings X.3 Condition (ii) requires that these consumption and
portfolio decisions be market clearing. Since the market is always cleared, the
consumer will never be observed except in the state z = 1. On the other hand, the
consumer has (though he always rejects it) the option to choose security holdings
x # 1. To evaluate these options, he needs to know v(z, y) for all Z.4
4. CONSTRUCTION OF THE EQUILIBRIUM
We begin by studying the consumer's maximum problem (i) for given price
behavior p(y). We have the following proposition.
PROPOSITION 1: For each continuous price function p( ) there is a unique,
bounded, continuous, nonnegative function v(z, y; p) satisfying (i). For each y,
v(z, y; p) is an increasing, concave function of z.
PROOF: Define the operator Tp on functions v (z, y) such that (i) is equivalent
to Tpv = v. The domain of Tp is the nonnegative orthant L 2n+ of the space L2n of
continuous, bounded functions u: E'+ x E+--> R, normed by
Ilull
=
sup lu(z,
Y)i.
z,y
Since applying Tp involves maximizing a continuous function over a compact set,
Tpu is well defined for any u e L Since U(c) is bounded, Tpu is bounded, and
by [2, p. 116] Tpu is continuous. Hence
T,:
L2n? L2n+.
T,
is monotone (u ,v
implies Tpu > T,pv) and for any constant A, T,p(u
+
A)
=
Tpu
+
PA.
Then from [3,
Theorem 5] Tp,
is a contraction mapping. It follows that Tpv
= v has a unique
solution v in L2n+ as was to be shown. Further, limn
Tun
=
v for
any
u E L2n+.
3
The bound z on x is to assure that the maximization in (i) is always over a compact set, even if
some components of p(y) are zero.
4This is not a "new" concept of equilibrium. It is (though no proof is offered) a standard,
Arrow-Debreu equilibrium where the commodity space is the space of all possible realizations of the
process
XYiyi.
See [12] for a full development of this relationship in a closely related context.
ASSET PRICES 1433
To prove that v is increasing in z, observe that Tpu is an increasing function of
z for any u. Since v = Tpv, this implies that v is increasing in z.
To prove that v is concave in z, we first show that if u (z, y) is concave in z, so
is (Tpu)(z, y). Let zo, z1 be chosen, let O 0
-
1, and let z9 =
Ozo+(1
-
0)zl. Let
(ci, xi) attain (Tpu)(z', y), i = O, 1. Now
(c@, x@)= (0c0+ (1- 0)c1, x0+
(1-0)x1)
satisfies c6 +p(y) x6
-
y*z' +p(y)*z6, so that
(Tpu)(z6, y)? U(c')+3 Ju(xO, y') dF (y', y)
-?: 0
(Tpu
Xz
, y )+
(1-
0)(Tpu
Xz1,
y)
using the concavity of U and u. Hence (Tpu)(z, y) is concave in z. It follows by
an induction that Tpu is concave in z for all n = 1, 2,. Then, since
lim,,.
Tpu = v, v is concave.
The derivatives of v with respect to z are described in the following
proposition.
PROPOSITION 2: If v(z, y; p) is attained at (c, x) with c >0, then v is differen-
tiable with respect to z at (z, y) and
(2) av(z,
Y; P)
=
U(c)[yi +pi(y)]
(i = 1,...,
n).
PROOF: Define f: R
+
-
R
+
by
f(A)==max
{
U(c)+f3Jv(x, y')dF (y',y)
c,x
subject to
c+p(y)-x-A, c,x 0.
For each A, f (A) is attained at c(A), x (A) say, and since the maximand is strictly
concave in c, c(A) is unique and varies continuously with A [2, p. 116]. If
c (A)> 0 and if h is sufficiently small, c (A)+ h is feasible at "income" A +
h, and
c (A
+
h)
- h is feasible at income A. Thus
f(A
+
h) u(c(A)+ h)+,3Jv(x(A), y')
dF
(y', y)
= u(c(A)+ h)- u(c(A))+ f(A)
and
f(A)> u(c(A +h)-h)+,4v(x(A
+h), y', y)
-
u(c(A
+
h)- h)- u(c(A
+
h))+f(A
+
h).
Combining these inequalities gives
U(c(A)+ h)- U(c(A))-< f(A
+
h)-f (A)
1434 ROBERT E. LUCAS, JR.
Dividing by h, letting h
-*
0, and utilizing the continuity of c ( ) gives
f'(A)= U'(c(A)).
Now letting A=y.z+p(y)-z, so that v(z,y;p)=f(A), we obtain (av/azi)=
f'(A)(aA/azi),
as was to be shown.
With the main features of v(z, y; p) thus established, we proceed to the study
of the maximum problem (i), still taking asset prices p to be described by an
arbitrary continuous function. The first order conditions, necessary and sufficient
in this instance, are:
(3) U(x(ypi(y)'=)a
(
Y)dF(y',y)
(i=,.
),
(4) c +p(y)
x =
y z
+p(y)
z,
provided c, x >0. If next period's optimum consumption c' is also positive,
Proposition 2 implies in addition
(5) v y )= U'(c')[y
+pi(y')] (iW=
1,..., n).
axi
Now in equilibrium (condition (ii)) z = x =
1,
c = I
yj,
and c' =
Xjy'.
Combin-
ing (3) and (5) and using these facts gives
(6)
U'(
y1)pi(y)
= U4 y'
y;)(y' +P(y'))
dF
(y', y),
for i =
1, . . . , n. One may think of (6), loosely, as equating the marginal rate of
substitution of current for future consumption to the market rate of trans-
formation, as given in the market rate of return on security i. Mathematically, (6)
is a stochastic Euler equation. It is conceptually the same as equations (8) in [10].
Since equation (6) does not involve the particular value function v(z, y; p)
used in its derivation, it must hold for any equilibrium price function. Con-
versely, if p*(y) solves (6) and v(z, y; p*) is as constructed in Proposition 1, then
the pair (p*(y), v(z, y; p*)) is an equilibrium. Thus solutions to (6) and equi-
librium price functions are coincident.
To study (6), define
gi(Y)=1{ U'( y') y
dF
(y', y) (i= 1,. ..., n).
Then if the n independent functional equations
(7) f(y)= gi(y)+ 3{(y')
dF
(y', y)
(i=1,. ., n)
have solutions (fi(y), ... ,
fn(y)),
the price functions
(8)
pi (Y) =
rL'i
. n),
ASSET PRICES 1435
will solve (6), and p(y)=
(P1(y),..
.,
p,(y))
will be the equilibrium price
function.
If f is any continuous, bounded, nonnegative function on En+,
the function
Tif:E-n+>R+ given by
(9) (Tif )(y)= gi(y)+ 3j'f(y')
dF
(y', y)
is well-defined and continuous in y. Since U is concave and bounded (by B, say)
we have for any c:
O
=
U
(O)
-
U
(c)
+ U'
(c)(- c)>-
B
-
cU'(c)
so that cU'(c)'< B for all c. It follows that the functions
gi(y)
are bounded, since
they are nonnegative and their sum is bounded by ,3B. Then the operators Ti
defined by (9) take elements of the space L of continuous, bounded functions
into the same space. Evidently, solutions to Tif =f are solutions to (7), and
conversely. We have, then, the following proposition.
PROPOSITION 3: There is exactly one continuous, bounded solution fi to (7) (or
to Tif
= f ). For any fo E Ln
+
limn _O Tnfo
=
fi.
The proof follows from the fact that Ti is a contraction, verified as in the proof of
Proposition 1.
In summary, we have learned that there is exactly one equilibrium price
function for this economy, and we have in (6) (equivalently in (7) and (8)) an
equation useful in characterizing it. In the next two sections, we develop further
results at this "general" level, and then turn to the study of the nature of
equilibrium prices in special cases.
5. A "DUALITY THEOREM"
There is a second way to construct the equilibrium price function, as will be
shown in this section. Since the preceding section already provides one way, this
method appears somewhat redundant in the present context. The second
method is slightly more general however (since it does not require differen-
tiability of U); it is also suggestive for stability theory.
Consider the functional equation
(10) r(z,y)= infn+ [sup U(c)+,
{
r(x, y)'dF(y', Y
qeE'~ C,X
subject to c +q- x y z +q z.
It will turn out that optimal policy functions q(z, y)
for this dynamic program
are, when evaluated at
(1, y), equivalent to the equilibrium price functions
found
in Section 4.
1436 ROBERT E. LUCAS, JR.
To study (10), let B be the space of bounded integrable functions on En+x
En, and let M: B
-*
B be the operator such that (10) is equivalent to: r = Mr.
For the record, we have the following proposition.
PROPOSITION 4: There is exactly one bounded integrable function r satisfying
r
=
Mr, and for any ul E B,
lim"""Mnu
= r.
The proof parallels that of Proposition 1, and will be omitted. In fact, much
more can be said about the function r.
PROPOSITION 5: The solution r to (10) satisfies
(11) r(z, y)= U(y z)+f3Jr(z, y')
dF
(y', y).
Further, r is continuous, and nondecreasing and concave in z for each fixed y.
PROOF: Define R: L2n +*L2n? by
(Rw)(z, y)= U(y z)+?3Jw(z, y')
dF
(y', y)
so that (I 1) reads: r Rr. We show that if w is continuous, and non-decreasing
and concave in z for each y, then (i) Rw has these properties, (ii) Mw
=
Rw.
The proof of (i) parallels arguments in the proof of Proposition 1, and can be
omitted.
To prove (ii), observe that the point (c,x)=(y z,z) satisfies c+q x
yz +qz for all q, so that Mw ?-Rw. Since w is concave, for any (z, y) the set
A =
(c, x): U(c)+?3
{
w(x, y') dF (y', y)?
(Rw)(z,y)
is convex. From the separation theorem for convex sets, there is a number
a(
and
a vector a
c En (not both zero) such that (c,x)c A implies aoc+ a x
a(y z + a z. Since U(c) is strictly increasing, it follows that
ao
> 0 and a a 0 0, so
we can define q
=
(a/ao) and write
(12) (c,x)eA implies c+q-x-y z+q z.
Now for this vector q, suppose there is a (c, x) in the interior of A with
c +?q x
=
y-? +q-z. Then by reducing c slightly, we obtain a point (c', x) in A
such that c'?+ q x < y z + q z: a contradiction to (12). This proves that q attains
Mw, or that Mw
=
Rw.
Finally, the properties listed for r follow easily from the fact that r solves (11),
using the methods applied to the proof of Proposition 1. This completes the
proof.
As immediate corollaries, we have the following propositions.
PROPOSITION 6: For all y,
r(q,
y)= v(1, y).
PROOF: From the definition of equilibrium v is the solution to (1 1) with z
=
1.
ASSET PRICES 1437
PROPOSITION 7: If p (y) is an equilibrium price function, then q (A, y) = p (y)
attains
rQ,
y).
The converse to Proposition 7 is the following.
PROPOSITION 8: If q (A, y) attains r(K, y) then p (y) = q (A, y) is an equilibrium
price
function.
PROOF: Let
q(Q,
y) attain r(l, y) and suppose that (co, xo) uniquely attains
(13) max
U(c)+,f3
r(x, y') dF (y', y)}
c,x
subject to
c
+q(1,
y) x
<
y 1 + q(1, y)* 1.
If (c , x ) =
(Xiyi,
1), then the assertion follows from Proposition 6 and the
definition of equilibrium. If (co, xo)# (iyi, 1), then a convex combination of
these two points is feasible for problem (13) and yields a higher value to the
objective function (since r is concave in z and U is strictly concave) contradict-
ing the assumption that (co, x?) solves problem (13).
6. STABILITY OF EQUILIBRIUM
The preceding sections showed that there is only one way for the economy
under study to be in competitive equilibrium: when all output is consumed, all
asset shares are held, and asset prices follow (6), or equivalently, solve the
dynamic program (10). As always, there are innumerable ways for the economy
to be out of equilibrium, so we must expect any treatment of out-of-equilibrium
behavior to have considerable arbitrariness, not resolvable by economic
reasoning. On the other hand, the model described above "assumes" that agents
know a great deal about the structure of the economy, and perform some
non-routine computations. It is in order to ask, then: will an economy with
agents armed with "sensible" rules-of-thumb, revising these rules from time to
time so as to claim observed rents, tend as time passes to behave as described in
Sections 4 and 5?
To sharpen this loosely posed question somewhat, let us recognize at least
three different stability questions raised by this model, and dispose of two of
them at once. First, in each period an ordinary "static" market clearing occurs, in
which current asset prices are set. Since stability in this sense is well understood,
we need add nothing here except the assumption that it always obtains. Second,
agents may be in ignorance of the distribution F(y', y) of the exogenous pro-
duction shocks, and learn its characteristics only gradually. Stability in this sense,
too, is a well understood problem in Bayesian decision theory [5, Ch. 10] and
need not be discussed here. Finally, consumers may be in error as to the price
function, or equivalently, about the distribution of future prices conditional on
1438 ROBERT E. LUCAS, JR.
the current state, or again equivalently, about the way they wish to evaluate their
end-of-period portfolio, x. We focus here on this last kind of disequilibrium.
The "correct" way, given preferences, to evaluate an end-of-period portfolio
x is to use the equilibrium value function v:
fv(x,
y') dF (y', y), but agents must
know this, and the economy must be in equilibrium for this valuation to be
correct. Suppose instead that agents use some other function u(z, y), say, where
u is continuous, concave, and increasing in z, but otherwise arbitrary, so that an
end-of-period portfolio x is valued at fu(x, y') dF (y', y). (To retain the con-
veniences of the representative consumer device, we are forced to treat all
agents as being wrong in the same way.) Suppose on the basis of this arbitrary
portfolio evaluation formula, asset demands are drawn and a current period
market clearing asset price vector q is es-tablished, at which c =
1i'i and x = 1.
Now if prices are established in this fashion, what will be the realized utility
yields experienced by agents?
The answer is given by the function (Mu)(z, y), where M is the operator
defined in association with equation (10). That this is so is the content of
Proposition 5: the price q which attains the right side of (10) is precisely that
price which clears markets, given the portfolio valuation function u.
If this experience is utilized by agents, they will replace the initial valuation u
with the value Mu actually experienced, then new prices will be established, and
utilities M2u experienced, and so on.5 Since, as shown in the preceding section,
M-u
-
v, where v is the equilibrium value function, prices will converge to the
equilibrium price function. In short, the successive approximations used in
Section 5 constitute a kind of stability theory.
It is worth emphasizing that the adjustment toward equilibrium described by
these successive approximations does not presuppose that agents are familiar
with the theory of Markov processes or of dynamic programming; nor need
agents in equilibrium be particularly skilled at responding to survey questions
about future price movements. All that is required is they have consistent
preferences for consumption and asset holdings (which would seem necessary for
dealing in asset markets at all) and that they revise these preferences in the
direction of the consumption utility actually yielded by their asset holdings.
The point of this section, it should also be said, is not that one would use any
of the successive approximations Mnu as a description of observed behavior.
(This suggestion is not even operational, since u was arbitrarily chosen.) It is
rather to argue that there is a theoretical reason for expecting the equilibrium to
be a good approximation to behavior. Certainly one would not expect to
capture
the creativity which is devoted to discovering and gaining
from
disequilibria in
actual economies in any mechanical approximation routine.
5As one of the referees for this paper emphasized, the process by which u is "replaced" by Mu,
Mu by M2u, and so forth, might well be quite complicated to spell out. It involves "learning" a
function over time by experiencing discrete values of the function Mu at arguments partly selected
by the household (z) and partly by nature (y). There are many ways to formulate learning of this sort;
for our purposes here, it seems simpler just to assume that households are good at it.
ASSET PRICES 1439
7. EXAMPLES
7.1. Linear Utility
The case of constant marginal utility of consumption does not exactly fit the
assumptions of Section 3 (it violates boundedness) but is easily handled
separately, and is a useful point of departure. In this case, equation (6) reduces
to
(14) pi(y)=f3E(y'Iy)+fE(pj(y')Iy)
which may be solved for
co
Pi (y )= L j3'E(yj,t+sjyjt
=
Y)-
S=1
That is, the price of the ith asset is the expected, discounted present value of its
real dividend stream, conditioned on current information y.
7.2. One Asset
It is easy to use equation (6) (or (7)) to characterize the function p(y), as can
be illustrated for the case of a one-asset economy. The crucial issues are the
information content of the current state y (that is, the way F(y', y) varies with y)
and the degree of "risk aversion" (the curvature of U). Suppose, as a first case,
that {Yt} is a sequence of independent random variables: F(y', y)= 1(y'). Then
g(y) is the constant
g
=
13{y'U'(y') dO (y')
=
13E[yU'(y)]
and calculating f from (9) as limn ,,TnO, say, we get
g
(Y)
-
PO ,f(y)=
0.
1 -:'
Then differentiating (8) gives
P3E[yyU'(y)]U"(y)
-U'(y)
P,") (1~)U()2 - U'(y)
0
Rearranging,
yp'(y) yU"(y)
p(y) U')
That is, the elasticity of price with respect to income is equal to the Arrow-Pratt
[1] measure of relative risk aversion.
In a period of high transitory income, then, agents attempt to distribute part
of
the windfall over future periods, via securities purchases.
This attempt
is frus-
trated (since storage is precluded) by an increase in asset prices.
1440 ROBERT E. LUCAS, JR.
Next, we consider autocorrelated production disturbances, under a restriction
which amounts to requiring that the stochastic difference equation governing Yt
have its root between zero and one: assume that F is differentiable, and that its
derivatives F1 and F2 satisfy
(15) 0<-F2 <F1.
We will repeatedly apply the following lemma.
LEMMA 1: Let F satisfy (15), and let h (y) have a derivative bounded between 0
and h' > 0. Then
(16)
--
I h(y')dF (y',y),h'.
dy
Jy
M
PROOF: Use the change of variable u = F(y', y), and invert to get y' = G(u, y),
so that G2= (-F2)/F1. Then the derivative in question is
-d h(G(u, y)) du = { h'(G)G2(u, y) du
dy
and the result follows from (15).
Now from (9), for any differentiable f,
(17)
d
MXy)
=
g'(Y)+ /3
d
fty')
dF
(y y)
dy
dy'
and from the definition of g(y),
(18) g'(y)=
3{
d
U'(y')y'
dF
(y, y).
To get any information on the slope of the solution f(y) to (7), then, we must
begin with bounds on the derivative of U'(y)y, or on U"(y)y
+
U'(y). (This
derivative is U'(y)[1
- R (y)], where R is the Arrow-Pratt measure of relative
risk aversion, so its magnitude has received some consideration.) For the sake of
discussion, take 0 and a as lower and upper bounds on U"(y)y + U'(y). Then
applying Lemma 1 to (18),
0
g, (y )
--
a
Then repeated application of (17), using Lemma 1 at each step, yields
(19) 0 f (Y _
where f(y) is the solution to (8) in this one asset case.6 From (8), the elasticity of
6Differentiability of the approximations TnF does not imply the differentiability
of
f,
and in
fact,
there is no easy way to verify this. For "f(y)< c" read: "f(y1)-f(yo)s c(y1
-
yo)."
ASSET PRICES 1441
the equilibrium price function is
(20)
yp(Y)=
yf'(y) yU"(y)
(2)
p(y)
f
f(y) -U'()
The second term on the right of (20) is the "income effect" we have seen
above; it is positive. The first term might be called the "information effect";7 it
has the sign of f'(y). Evidently, the use one can make of these formulas depends
on our knowledge of the curvature of U; (19) and (20) show how to translate
such knowledge into knowledge about asset prices.
In the present case of relative risk aversion8 less than unity, we have found in
(19) that f'(y)> 0, so that the information effect is positive. Thus as one might
expect, new optimistic information on future dividends leads to increased asset
prices. (Of course, one might also expect that this information will lead to an
attempted consumption binge now, lowering asset prices!)
Observationally, the derivative p'(y) is the change in the ratio of a compre-
hensive stock price index to the CPI, as real output varies. Even in the simplified
economy under study, then, the relationship of asset prices to real output is far
from simple and possibly not even monotonic. Perhaps it has been good judg-
ment, not merely timidity, which has led aggregate theorists to steer clear of any
attempt to "understand the market."
7.3. Many, Independent Assets
If the number of productive units is large, and if there is sufficient indepen-
dence across units, one would expect that replacing the term
U'(Xiyi)
in (6) with
U'(A),
where
A =EAi=ZJyi(y)dy
i i
in mean total output, would yield a good approximation to the
equilibrium price
function. Let us pursue this idea, and the question
of
approximation generally,
with the aid of the next lemma.
LEMMA 2: Let S, T: L
-+
L be contractions with modulus f3 and fixed points fs,
fT E L. Suppose that
jSf -Tfjj
A
for all f
e L.
Then
lIfs -fTII 1
A-
.
7This follows Grossman
[8].
8
In this multiperiod context, the term "risk aversion" is perhaps misleading, since the curvature of
U also governs the intertemporal substitutability of consumption. With time-additive utility, there is
no way to disentangle these conceptually distinct aspects of preferences.
1442 ROBERT E. LUCAS, JR.
PROOF: For any f,
jIS2f
-
T2fj
1
IIS2f
-
TSf
11
+
|ITSf
-
T2fjj
I|S(Sf )-T(Sf )1I
+
f 3lSf- Tfl
SA+
PA,
and, in general,
ljSnf-TTnfjiA(1+3+.
+fln-1)
Letting n
-+ x gives the result.
Now if gi(y) is an approximation to
gi(y),
and T is defined by
Tif
=
gi(y)+.3Jf(y')
dF
(y', y),
we have
I?Tif- Titff = lgi (Y)-gi(Y)lI
Then if
fi
and
fi
are the fixed points of Ti and Ti, respectively, Lemma 2 gives the
bound
(21)
-fill-f<
(1-j)-1jj-i -gill.
Returning to the specific approximation proposed above, let
gi(y)
=
U'(u
JYi
dF
(y', y),
define Ti
as above, and let
fi
be the fixed point of
Ti.
Then the approximate price
function
f i(Y)
pii(Y)=
is just the solution calculated in 7.1 above.
To evaluate this approximation, we need bounds on
gi
-IgI.
To this
end,
let
us bound U"(y):
11 U"I'
--M. Then
gi(y)-
g()I =
fI|J[U' E; y
-U'(A()]y
dF
(y,
Y)
EyH11(
Y-Yi'dF (Y' Y)|
using the mean value theorem. If the
yi's
are
independent,
or if
F(y', y)=
HkFk(yk, Yk),
then
J(~
y
i-LY
dF (y', y)=
-(
(Yi)))(Y-'- dF (y' y)
+
pj
E (y;'- )
dF
(y', y)
-var
(y
i'l yi) +Ai
E
E(y;-yj).
ASSET PRICES 1443
Combining gives
jgi,(y)-g) ) sup [var (y 'yi)+ A'
Z
E(yX
-
Ailyi)].
y
i
If we think of a sequence of economies of the same total size, but with more and
more independent productive units, of roughly equal size, var
(y jyi)
and
Ai
-
plYi) will
tend to zero, and the
approximations fi
will become close.
8. THE MARTINGALE PROPERTY
We have shown that equation (6) exhausts the implications of the assumption
that, in this model economy, prices are in equilibrium and "reflect all available
information." Evidently, asset prices themselves do not possess the Martingale
property. The series that does have this property (something has to, in this
time-additive set up) is the series
wi,(i
= 1, ... , n) defined by
Wi-t+i-Wit
=
lu(z yit+1)(Yi,t+ +Pi,t+i)-
(U
Yj,)Pit,
since from (6), the expectation of the right side of (22), conditioned on all
available information (in this case, Yt) is zero.
If the terms
U'(Iiy1t)
do not vary much, either because agents are indifferent
-to risk (example 7.1) or because there is little aggregate risk (example 7.3), then
securities prices properly "corrected" for dividends Yit almost have the property
but not without another "correction" for the discount factor
jl.
In any case,
neither rationale for a constant U'(X1y1t) seems likely to closely approximate
reality.
It should be added that the importance of the requirement that "the conditions
of market equilibrium can be stated in terms of expected returns" has been
repeatedly emphasized by Fama and other efficient market theorists; it is not a
new result from this paper. What is new, I think, is an explicit framework within
which one can judge what this requirement means and whether or not it is
satisfied, or which in other words can lend some insight into the conditions under
which the Martingale property is likely to approximately describe a price series.
Within this framework, it is clear that the presence of a diminishing marginal
rate of substitution of future for current consumption is inconsistent with this
property.9
9. CONCLUSIONS
What can be concluded from this exercise (beyond the observation that a little
knowledge of geometric series goes a long way, or perhaps, is a dangerous
thing)? Substantively, the discussion of stability of Section 6 indicates that the
9
This complements Danthine's [4] finding that a diminishing marginal rate of transformation over
time, in a model with storage, has the same effect.
1444 ROBERT E. LUCAS, JR.
applicability of the hypothesis that agents "know" the "true" probability dis-
tributions of future prices has little to do with the question of whether agents
(ourselves included) think of, or describe, their behavior in these terms. A
relatively crude use of hindsight, applied in a reasonably stationary physical
environment, will lead to behavior well-approximated by rational expectations.
With respect to the random character of stock prices, it is evident that one can
construct rigorous economic models in which price series have this charac-
teristic10 and ones with equally rational and well-informed agents in which they
do not. This would suggest that the outcomes of tests as to whether actual price
series have the Martingale property do not in themselves shed light on the
generally posed issue of market "efficiency."
In the main, however, this paper is primarily methodological: an illustration of
the use of some methods which may help to bring financial and economic
theories closer together. It may help, then, to close with some guesses as to the
fronts on which further progress can be expected.
The time-additive preference structure is, as remarked earlier, a nuisance, and
it has no rationale beyond tractability. It would not be difficult (with the aid of
[6]) to use recursive, but non-additive preferences of the Koopmans-Diamond-
Williamson [9] type, provided sufficient "impatience" is assumed.
Second, one would like to introduce capital accumulation. In this regard, the
marginal analysis of Section 4 is probably a dead-end: equation (6) is a kind of
Euler condition, and will necessarily involve capital provided capital enters the
model in a non-trviial way. Aside from special cases (such as the one studied in
Section 4) stochastic Euler equations are not likely to be of value in constructing
solutions. Equation (10) in Section 5 appears more promising; perhaps it has
useful analogues in more generally formulated models.
University of Chicago
Manuscript received September, 1975; final revision received March, 1978.
REFERENCES
[1] ARROW, KENNETH J.: Aspects of the Theory of Risk Bearing. Helsinki: Yrjo Jahnsson
Lectures, 1965.
[21 BERGE, CLAUDE: Topological Spaces. New York: Macmillan, 1963.
[3] BLACKWELL, DAVID: "Discounted Dynamic Programming," Annals of Mathematical Statis-
tics, 36 (1965), 226-235.
[4] DANTHINE, JEAN-PIERRE: "On the Relevance of the Efficient Market Model," Carnegie-
Mellon University Working Paper, 1974.
[5] DE GROOT, MORRIS H.: Optical Statistical Decisions. New York: McGraw-Hill, 1969.
[6] DENARDO, ERIC V.: "Contraction Mappings in the Theory Underlying Dynamic Pro-
gramming," SIAM Review, 9 (1967), 165-177.
[7] FAMA, EUGENE F.: "Efficient Capital Markets: A Review of Theory and Empirical Work,"
Journal of Finance, 25 (1970), 387-417.
[8] GROSSMAN, SANDFORD: "The Existence of Futures Markets, Noisy Expectations, and
Informational Externalities," unpublished University of Chicago Working Paper, 1975.
10
Samuelson did this in [14].
ASSET PRICES 1445
[9] KOOPMANS, TJALLING C., PETER A. DIAMOND, AND RICHARD E. WILLIAMSON: "Sta-
tionary Utility and Time Perspective," Econometrica, 32 (1964), 82-100.
[10] LEROY, STEPHEN F.: "Risk Aversion and the Martingale Property of Stock Prices," Inter-
national Economic Review, 14 (1973), 436-446.
[11] "The Determination of Stock Prices," unpublished University of Pennsylvania
Doctoral Dissertation, 1971.
[12] LUCAS, ROBERT E., JR., AND EDWARD C. PRESCOTT: "Investment under Uncertainty,"
Econometrica, 39 (1971), 659-681.
[13] MUTH, JOHN F.: "Rational Expectations and the Theory of Price Movements," Econometrica,
29 (1961), 1-23.
[14] SAMUELSON, PAUL A.: "Proof that Properly Anticipated Prices Fluctuate Randomly,"
Industrial Management Review, 6 (1965), 41-49.